NVIDIA held its 2026 Annual General Meeting on June 24, where shareholders approved all 10 board nominees. CEO Jensen Huang declared that the “Useful AI” era has arrived — and that AI infrastructure is transitioning from the experimental phase to full production, with a build cycle that will span decades.
Huang noted that roughly 30 million software developers worldwide collectively earn about $3 trillion annually, supporting an estimated $100 trillion in global economic activity. With the rise of Agentic AI, these values are being further amplified — and the question of AI ROI, he said, “already has an answer.” A growing number of industries are accelerating adoption of agentic AI.
🧩 The AI “five-layer cake”
Huang compared the AI industry to a five‑tier stack: energy, chips & systems, infrastructure, models, and applications. While traditional data centres provide compute and storage, today’s AI factories produce tokens — and each token represents profit. “Useful AI” is profitable AI, and that is the core reason global compute demand remains persistently strong.
Huang emphasised that while NVIDIA systems may not be the lowest‑cost option upfront, they deliver the lowest token‑generation cost, highest token throughput, and highest commercial revenue — making them the most economical choice in the long run.
Inference overtakes training · performance per watt
Addressing the growing view that inference demand may eventually surpass training, Huang said that as inference becomes the core of AI commercialisation, the key competitive metric for future AI infrastructure will be performance per watt. Products based on the Blackwell architecture, he said, have already given NVIDIA a leading edge in inference.
📡 Roadmap: Hopper → Blackwell → Vera Rubin
Hopper was designed primarily for pre‑training. Blackwell extended reasoning capability to rack‑scale systems. The next‑generation Vera Rubin is purpose‑built for agentic AI and is now in full production. Huang revealed that nearly all major model developers, public cloud providers, and hyperscalers are already preparing deployments based on Vera Rubin.
The Vera CPU has begun receiving orders, and Huang said it could become “one of the most important products in NVIDIA’s history.”
“Largest infrastructure build in human history”
On the longevity of AI infrastructure spending, Huang argued that more compute generates more tokens, which drives higher revenue. This infrastructure cycle will be measured in decades, involving upgrades to power grids, internet backbones, and critical systems. “This will become the largest infrastructure build in human history,” he said.
🤖 Physical AI · next growth wave
Looking ahead, Huang sees Physical AI as NVIDIA’s next major growth opportunity. Robots, autonomous vehicles, and smart factories will become real‑world AI agents — capable of perception, reasoning, planning, and execution — which will in turn drive another wave of infrastructure investment.
Shareholder returns · 50%+ free cash flow
On capital returns, Huang confirmed that based on confidence in continued market growth and free‑cash‑flow generation, NVIDIA plans to return 50% or more of its free cash flow to shareholders this year, next year, and over the long term. The company has already authorised an $80 billion share buyback and increased its dividend.